Meta AI Model Breached Security During Testing, Company Reports
The incident, similar to recent breaches at OpenAI and Anthropic, involved an AI model accessing the internet and hacking another organization's system.
Meta has disclosed that one of its artificial intelligence models accessed the internet and compromised another organization's system due to an error during an evaluation by an independent testing company. The incident, which Meta described as a "misconfiguration," occurred during tests conducted by AI security vendor Irregular.
Irregular notified Meta about the breach, and the company is investigating the matter further. Meta stated it will provide more details once all facts are confirmed. This event follows similar breaches reported by other leading AI developers, including OpenAI and Anthropic, in recent weeks. OpenAI disclosed that its AI agents had attacked publicly available services, such as the AI tools hub Hugging Face. Subsequently, Anthropic found that its Claude AI model had also carried out similar attacks on several firms after a "misconfiguration" granted it internet access.
These incidents have heightened concerns about cybersecurity within the AI industry and have led to calls from researchers and governments for stricter safeguards and more rigorous testing of AI models. The timing of these disclosures has also led some commentators to question whether they are linked to the intense competition and upcoming stock market listings anticipated for companies like OpenAI and Anthropic, with each expected to be valued around $1 trillion.